| 總結: | Accurate and efficient human pose estimation is crucial for precise motion tracking and performance feedback in real-time sports analysis. This paper presents a real-time pose estimation framework that integrates EfficientPose and T-GCN (Temporal Graph Convolutional Networks) to address the challenges of dynamic and complex sports scenarios. EfficientPose utilizes a highly efficient network architecture to achieve accurate 3D pose estimation from single-frame images, providing a robust foundation for subsequent temporal modeling. T-GCN further refines the motion trajectories by modeling temporal dependencies and spatial relationships across frames, ensuring temporal continuity and spatial consistency. Experimental results demonstrate the superior performance of the proposed framework, achieving the lowest Mean Absolute Error (MAE: 30.5 mm), the highest Multiple Object Tracking Accuracy (MOTA: 80.2%), and maintaining a real-time frame rate (45 FPS) across multiple benchmarks. Compared to traditional methods, the proposed approach exhibits significant advantages in handling high-speed motions, occlusions, and complex multi-agent interactions, enabling high-precision and temporally stable pose estimation. This framework provides an efficient and robust solution for real-time sports performance analysis, offering valuable scientific support for performance feedback and tactical decision-making.
|
|---|
